Although many people talk about autonomous driving as if it’s something impossible, driverless cars are already here. Tesla’s upcoming robotaxi unveiling on October 10 looks exciting, but it’s unclear if it will change the autonomous vehicle landscape.
Others point out that Tesla’s FSD is not autonomous driving because a driver needs to “curate” every FSD action, allowing it or taking over the steering wheel to correct it. Even though Tesla’s FSD has accumulated over a billion miles, the fact that a driver still needs to supervise it makes it a mere Level-2 autonomous driving system. I know this infuriates Tesla fans, who consider Tesla FSD superior to all Level-3 systems currently certified, including those from Mercedes-Benz and BMW.
Is Tesla autonomous or what?
Although it’s unclear whether he still stands behind this statement, Elon Musk announced in 2016 that all Tesla EVs have all the hardware required to become autonomous one day when the software is ready. This made a lot of people believe that Tesla could turn all its cars into Level 5 autonomous vehicles “with the flip of a switch.” Being Level 2 is just semantics for them. What matters is that Tesla FSD could drive everywhere, whereas these certified Level-3 systems require special conditions (and locations) to work.
However, this is how SAE classifies autonomous driving systems, and it’s not going to change until Tesla is willing to assume responsibility for whatever FSD decides during driving. Tesla acknowledged that when it changed the wording on its website and admitted the FSD software is a Level 2 autonomous driving system. This means that Tesla doesn’t want to be held responsible for whatever FSD might do wrong and prefers to blame it on the drivers.
Seen through this lens, Tesla FSD is not technically autonomous driving. Not only because Tesla doesn’t assume liability for FSD actions, but also because not even Tesla trusts it to do the right thing consistently. The EV maker makes this very clear in the FSD description and the disclaimers Tesla owners must agree to when they activate the self-driving features.
How about FSD safety?
Tesla FSD is still far from being capable of driving without supervision, even though many swear by it. Sure, for some, it may feel like magic to see the steering wheel turning by itself and the car acting like a human driver most of the time. However, as others are well aware, Tesla FSD might run a red light when you least expect it or do other stupid things. In some cases, it might put you, your family, and others in danger, something that Tesla warns about.
Currently, the number of miles between interventions seems to be Tesla’s preferred metric for FSD appraisal. According to the crowd-sourced data portal Tesla FSD Tracker, the latest FSD builds achieve about 134 miles between critical interventions or 78 miles between all kinds of interventions. According to the portal, critical interventions are safety-related, including those needed to avoid an accident, going through a red light/stop sign, being on the wrong side of the road, or other unsafe actions.
This means that, left to its own devices, the FSD software could get itself (and you) into trouble every 134 miles, give or take. That’s not human-like, as some claim, because most humans are not likely to cause a crash every 134 miles. Although Tesla data released in 2023 shows that FSD only recorded one crash every 3.2 million miles, that was precisely because a human driver was there to correct mistakes and avoid thousands of potential crashes.
Is Tesla close to achieving full autonomy?
Despite being bullish on the chances that FSD will become a Level 4 or even Level 5 system in the future, the current state of affairs doesn’t make me very optimistic. It’s even stranger that Tesla doesn’t seem more optimistic either. Recently, the AI team released an FSD roadmap, and nothing on it looked like it sees autonomous driving as possible.
The next iteration of Tesla Full Self-Driving promises a six-time improvement in miles between interventions. The roadmap doesn’t mention what kind of interventions are considered, but I assume critical interventions, which are less common. By this metric, we must expect about 800 miles between critical interventions with FSD V13. Even if it sounds like a massive improvement, it’s not enough to become safer than humans.
However, even though FSD V13 will start tests in October, as the Tesla AI team mentioned, it doesn’t necessarily mean it will be production-ready soon. FSD V12 was released to testers in January but only started customer rollout in March. Eventually, Elon Musk considered it so good that it changed the entire Tesla strategy around it. Since then, FSD V12 has seen a lot of hiccups and delays, while many owners have reported many issues.
Currently, Tesla is rolling out FSD V12.5.2 to customer vehicles, and the initial impressions don’t suggest that a robotaxi might be possible, even less so in a month. Even with the improvement announced for the V13, Tesla Full Self-Driving is not good enough to drive people around. This is probably why Tesla never applied for a robotaxi license.
What’s the competition doing?
While Tesla continues to work on its FSD software, other companies have already deployed driverless cars in several cities. Cruise and Waymo are among the most advanced, having started commercial operations. Both companies are expanding their service areas in the cities they operate in while entering new territories.
There’s a fundamental difference between Cruise and Waymo on the one hand and Tesla on the other. Unlike Tesla, which only uses cameras for its autonomous driving system, Cruise and Waymo fitted their autonomous vehicles with various sensors, including cameras, radars, and lidars. This approach provides for redundancy in case some sensors are disabled, such as cameras affected by rain or dirt.
The downside is that the vehicles become crazy expensive. This means that Waymo and Cruise must spend a lot of money to deploy a meaningful number of robotaxis, whereas Tesla already has millions that are much more reasonably priced. This would allow Tesla to scale its robotaxi business much faster once all regulatory hurdles are removed, and its FSD software gets safe enough.
Uber, which was once working on its own robotaxi project, seems to have abandoned the idea after one of its prototypes struck and killed a pedestrian in 2018. Instead, the ride-hailing company partnered with Cruise to employ its robotaxis on its platform. Ironically, Cruise was also involved in a deadly crash last year, causing a suspension of its license in San Francisco. Even more surprising, Cruise’s parent GM is an investor in Uber’s main rival, Lyft.
This partnership looks like a win-win for both companies. Uber has a huge customer base, while Cruise has the autonomous vehicles that Uber hasn’t stopped dreaming of. However, this partnership would not last long after GM cuts the funding, considering the high costs of every robotaxi. Uber’s business model is to squeeze drivers as much as possible, keeping the bulk of the profits to itself. It’s unclear how Cruise could compete with the low costs Uber benefits from employing human drivers.
Waymo, on the other hand, relies on the large coffers of its parent company, Alphabet, which also owns Google, among other things. Even though Waymo’s record hasn’t been stellar, and its beginnings looked less surefooted, it has become an autonomous driving powerhouse. Waymo operates a fleet of roughly 700 vehicles in San Francisco and Phoenix and has started operations in Los Angeles and Austin. It recently reported that it provides over 100,000 weekly paid rides.
The costs have also become more affordable, sometimes even below Uber. Many people prefer to use a Waymo where available, specifically because they do not want to deal with a driver. Many consider Waymo much safer, showing we’ve gone a long way in acceptance of autonomous driving. Besides the human factor, the price alone should give Uber the chills. Even if it introduces driverless rides thanks to its Cruise partnership, it would be interesting to see how both companies could compete on price with Waymo.
Autonomous driving competition is heating up
What’s impressive about Waymo is the speed at which it scales its business across different cities. Even though Tesla fans criticized it for not being entirely autonomous since it relies on high-definition mapping in geofenced areas, Waymo is constantly expanding those areas to entire cities. This is all the more impressive, considering that many thought self-driving cars were impossible. And Waymo not only does it, but it’s also safer than humans, as its safety data shows.
This was something that Tesla had always wanted, although it still needs to prove it. Waymo recently launched a data portal showing how its robotaxis compares to human drivers in the cities it operates in. Not only that, but the methods used to gather this data are transparent, establishing a new standard in the autonomous driving industry. The latest data available shows that Waymo robotaxis are about four times safer than humans, and this will only improve.
While autonomous driving might still take a few years to enter the mainstream, there’s no turning back now. It has already been proven that cars can drive without human supervision, and all we have to do is scale this to the extreme. It’s unclear for now which approach will yield the best results, Tesla Vision-only or Waymo’s “everything plus the kitchen sink” sensor array. Currently, both systems have their proponents, not only in the US but also in China.
This shows that Tesla, once considered a pioneer of autonomous driving, is already facing stiff competition when it enters the market. Tesla may have sabotaged itself by insisting that vision is better than a balanced set of sensors, including lidar. Not necessarily because autonomous driving would not work on vision only, but because it had to start everything from scratch. It then did it again with the switch to end-to-end AI, even though that was a necessary move.
Tesla invested billions in new AI data centers to train its neural networks for autonomous driving, and it’s confident that its solution will allow it to scale faster than rivals. Although this might be true in the future, last year’s data shows that improvement has been incremental, if anything. Even Tesla’s metric, miles between interventions, appears to have stagnated, and it’s unclear how it will take off from where it sits now.
Tesla robotaxi: what to expect from the October 10 event?
We’re one month away from Tesla’s big unveiling, and it won’t be long until we see what it has in store. The moment appears to have been rushed by an angry Elon Musk after Tesla employees leaked classified information to Reuters. In April, Musk announced the robotaxi would launch on August 8 but soon realized that Tesla could not deliver it by then. The event has been delayed until October 10, and preparations suggest that Tesla aims for the sky.
The EV maker booked Warner Bros. Studios in Burbank, California, as the venue for the event, which could mean several things. The studios can provide an environment that imitates a suburban neighborhood, perfect for robotaxi rides in a controlled area. Since August, Tesla has been gathering data in the region, suggesting that it has modeled the studios’ streets and houses into its database.
Currently, Tesla holds a permit to test autonomous vehicles, but only with a safety driver behind the wheel. It’s unclear whether this also applies to private areas. In any case, we should expect that people attending the event will get the chance to ride in robotaxi prototypes, with or without a safety driver.
However, Tesla might want to wow the attendees with more products and tricks, as Musk suggested when he said that Tesla would show “a few other things.” It’s unclear what these surprises might be, but I am certain everyone would be disappointed without them.
Based on some products Tesla has been working on, we could expect the robotaxi to feature wireless charging and show it off during the event. We could also see Optimus robots wandering around during test drives, allowing Tesla to show off the robotaxis interacting with pedestrians without putting anyone in danger.